CQL: A Language for Continuous Queries over Streams and Relations Jennifer Widom Stanford University Joint work with Arvind Arasu & Shivnath Babu stanfordstreamdatamanager

CQL: A Language for Continuous CQL: A Language for Continuous Queries over Streams and RelationsQueries over Streams and Relations

Jennifer Widom

Stanford UniversityStanford University

Joint work with Arvind Arasu & Shivnath Babu

stanfordstreamdatamanager

stanfordstreamdatamanager 2

Data StreamsData Streams• Continuous, unbounded, rapid, time-varying

streams of data elements

• Occur in a variety of modern applications– Network monitoring and traffic engineering– Sensor networks, RFID tags– Telecom call records– Financial applications– Web logs and click-streams– Manufacturing processes

• DSMSDSMS = Data Stream Management System


DBMS versus DSMSDBMS versus DSMS• Persistent dataPersistent data

• One-time queriesOne-time queries

• Random access

• Access plan determined by query processor and physical DB design

• Transient data Transient data (and persistent data)(and persistent data)

• Continuous queriesContinuous queries

• Sequential access

• Unpredictable data characteristics and arrival patterns


DSMS

Scratch Store

The (Simplified) Big PictureThe (Simplified) Big Picture

Input streams

RegisterQuery

StreamedResult

StoredResult

Archive

StoredData


(Simplified) Network Monitoring(Simplified) Network Monitoring

RegisterMonitoring

Queries

DSMS

Scratch Store

Network measurements,Packet traces

IntrusionWarnings

OnlinePerformance

Metrics

Archive

LookupTables


STREAMSTREAMThe Stanford Stream Data ManagerThe Stanford Stream Data Manager

• General-purpose DSMS for streams and stored data

• Relational (unlikely to change)

• Centralized server model (likely to change)

– Single-threaded and parallel versions

• Declarative language for registering Declarative language for registering continuous queriescontinuous queries


The STREAM System:The STREAM System:Some Implementation IssuesSome Implementation Issues

• Designed to cope with:– Stream ratesStream rates that may be high,variable, bursty

– Continuous query loadsContinuous query loads that may be high, volatile

• Primary coping techniques– Continuous self-monitoringself-monitoring and reoptimizationreoptimization

– Graceful approximationapproximation as necessary

– Careful resource allocation and useresource allocation and use

These issues do not affect the query language directly


The STREAM System:The STREAM System:Current StatusCurrent Status

• Version 1.0 up and running

• Implements most of query language

• Now incorporating the fun stuff

• Currently approx. 25,000 lines of C++ code

• Release plans

1) Web-accessible running server

2) Downloadable software


The Stream Systems LandscapeThe Stream Systems Landscape

• (At least) three general-purpose DSMS prototypes underway– STREAMSTREAM (Stanford)

– AuroraAurora (Brown, Brandeis, MIT)

– TelegraphCQ TelegraphCQ (Berkeley)

• Cooperating to develop stream system stream system benchmarkbenchmark One goal: demonstrate that conventional systems

are far inferior for data stream applications


Declarative Language for Declarative Language for Continuous QueriesContinuous Queries

• A distinction between Aurora and the other two systems (STREAM and TelegraphCQ)

– Aurora users directly manipulate a single large execution plan through a “boxes-and-arrows” interface

– STREAM compiles declarative queries into individual plans, system may merge plans

– STREAM also supports direct entry of plans


Declarative Language (cont’d)Declarative Language (cont’d)

• STREAM’s declarative language is based on SQLSQL syntax + some aspects of semantics– TelegraphCQ’s also based on SQL but more

loosely

• Modifications to standard SQL semantics for:– Streams plus relationsStreams plus relations

– Continuous query resultsContinuous query results

• Additional constructs for sliding windowssliding windows and samplingsampling


Aside: Query Language SemanticsAside: Query Language Semantics

• Around 1980

People were inventing, understanding, and implementing relational query languages

• Around 1990 People were inventing and implementing trigger

(active rule) systems without completely defining or understanding rule behavior

• Around 2000 Ditto for continuous queries


Aside on Semantics (cont’d)Aside on Semantics (cont’d)

• The semantics of SQL queries is (relatively) easy to understand– Even lots of SQL queries running together

• The semantics of a single trigger is (relatively) easy to understand– But lots of triggers together can be complex

• The semantics of even a single continuous query may not be obvious– But lots running together is no harder


A Nonobvious Continuous QueryA Nonobvious Continuous Query

• Stream of stock quotes: Stocks(ticker,price)

• Monitor last 10 minutes of quotes:

Select From Stocks [Range 10 minutes]

• Is result a relation, a stream, or something else?

• If a relation, what exactly does it contain?

• If a stream, how does query differ from:Select From Stocks [Range 1 minute]or Select From Stocks []


Another Nonobvious CQAnother Nonobvious CQ

• Stream of ordered items, table of item prices

• Prices for five most recent ordered items:Select P.priceFrom Items [Rows 5] I, PriceTable PWhere I.itemID = P.itemID

• Is result a stream or a relation?

• What if item price changes?


Our Original Working SemanticsOur Original Working Semantics

“The result of a continuous query at time T is the result of treating the streams up to T as relations and evaluating the query using standard relational semantics.”

+ Relies on existing well-understood semantics

– Asymmetric: streams-to-relations but not relations-to-streams

– Can break down for complex queries (subqueries, aggregation)


Our Final Semantics and LanguageOur Final Semantics and Language

• Abstract: Abstract: interpretation for continuous queries based on certain “black boxes”

• Concrete: Concrete: SQL-based instantiation for our system; includes syntactic shortcuts, defaults, equivalences


Goals in Language DesignGoals in Language Design

1) Support continuous queries over multiple streams and relations

2) Exploit relational semantics to the extent possible

3) Easy queries should be easy to write

4) Simple queries should do what you expect


Example Query 1Example Query 1Two streams, contrived for ease of examples:

Orders (orderID, customer, cost) Fulfillments (orderID, clerk)

Total cost of orders fulfilled over the last day by clerk “Sue” for customer “Joe”

Select Sum(O.cost)From Orders O, Fulfillments F [Range 1 Day]Where O.orderID = F.orderID And F.clerk = “Sue” And O.customer = “Joe”


Example Query 2Example Query 2

Using a 10% sample of the Fulfillments stream, take the 5 most recent fulfillments for each clerk and return the maximum cost

Select F.clerk, Max(O.cost)From Orders O, Fulfillments F [Partition By clerk Rows 5] 10% SampleWhere O.orderID = F.orderIDGroup By F.clerk


Formalization (and Rest of Talk)Formalization (and Rest of Talk)

• Formal definitions for relations and streams

• Formal conversions between them

• Abstract semantics

• Concrete language: CQL

• Syntactic defaults and shortcuts

• Equivalence-based transformations

• Benchmark, system screenshots


Relations and StreamsRelations and Streams

• Assume global, discrete, ordered time domain (more on this later)

• Relation– Maps time T to set-of-tuples R

• Stream– Set of (tuple,timestamp) elements


ConversionsConversions

Streams Relations

Window specification

Special operators:Istream, Dstream, Rstream

Any relational query language


Conversion DefinitionsConversion Definitions

• Stream-to-relation– S [W ] is a relation — at time T it contains all tuples

in window W applied to stream S up to T

– When W = , contains all tuples in stream S up to T

• Relation-to-stream– Istream(R) contains all (r,T ) where rR at time T

but rR at time T–1

– Dstream(R) contains all (r,T ) where rR at time T–1 but rR at time T

– Rstream(R) contains all (r,T ) where rR at time T


Abstract SemanticsAbstract Semantics

• Take any relational query language

• Can reference streams in place of relations– But must convert to relations using any window

specification language ( default window = [] )

• Can convert relations to streams– For streamed results

– For windows over relations (note: converts back to relation)


Query Result at Time Query Result at Time TT

• Use all relations at time T

• Use all streams up to T, converted to relations

• Compute relational result

• Convert result to streams if desired


Abstract Semantics – Example 1Abstract Semantics – Example 1Select F.clerk, Max(O.cost)From O [], F [Rows 1000]Where O.orderID = F.orderIDGroup By F.clerk

• Maximum-cost order fulfilled by each clerk in last 1000 fulfillments


Abstract Semantics – Example 1Abstract Semantics – Example 1Select F.clerk, Max(O.cost)From O [], F [Rows 1000]Where O.orderID = F.orderIDGroup By F.clerk

• At time T: entire stream O and last 1000 tuples of F as relations

• Evaluate query, update result relation at T


Abstract Semantics – Example 1Abstract Semantics – Example 1Select IstreamIstream(F.clerk, Max(O.cost))From O [], F [Rows 1000]Where O.orderID = F.orderIDGroup By F.clerk

• At time T: entire stream O and last 1000 tuples of F as relations

• Evaluate query, update result relation at T

• Streamed result:Streamed result: New element (<clerk,max>,T) whenever <clerk,max> changes from T–1


Abstract Semantics – Example 2Abstract Semantics – Example 2

Relation CurPrice(stock, price)

Select stock, Avg(price) From Istream(CurPrice) [Range 1 Day] Group By stock

• Average price over last day for each stock


Abstract Semantics – Example 2Abstract Semantics – Example 2

Relation CurPrice(stock, price)

Select stock, Avg(price) From Istream(CurPrice) [Range 1 Day] Group By stock

• Istream provides history of CurPrice

• Window on history, back to relation, group and aggregate


Digression: TimeDigression: Time• Relation updates carry timestamps too

• “Well-behaved” streams and relation updates:• Arrive in order

• No “delayed activity” – no long pauses then resume with very old timestamps

• Semantics of query language tacitly assumes well-behaved streams and updates• Poor behavior handled separately, not within

language

• Distinction from Aurora, which does not separate


Time Digression (cont’d)Time Digression (cont’d)

• Simplest scenario: global system clock– Stream elements and relation updates

timestamped on entry to system

– Guaranteed ordering, no delayed activity

– Query results also based on system time


Flexible Application-Defined TimeFlexible Application-Defined Time• Streams and relation updates contain

application timestamps– May arrive out of order

– May pause and resume

• Query results in application time

• Application generates “heartbeats”– Or deduce heartbeat from parameters: scrambling,

skew, latency, clock progress

– Reminder: separate from query language semantics


Concrete Language – CQLConcrete Language – CQL

• Relational query language: SQL

• Window specification language derived from SQL-99– Tuple-based windows

– Time-based windows

– Partitioned windows

• Simple “X% Sample” construct– May incorporate emerging SQL standard


CQL (cont’d)CQL (cont’d)

• Syntactic shortcuts and defaults– So easy queries are easy to write and simple

queries do what you expect

• Equivalences– Basis for query-rewrite optimizations

– Includes all relational equivalences, plus new stream-based ones

• Examples: already seen some, more later


Shortcuts and DefaultsShortcuts and Defaults

• Prevalent stream-relation conversions can make some queries cumbersome– Easy queries should be easy to write

• Two defaults:– Omitted window:Omitted window: Default []

– Omitted relation-to-stream operator:Omitted relation-to-stream operator: Default Istream operator on:

• Monotonic outermost queries• Monotonic subqueries with windows


The Simplest CQL QueryThe Simplest CQL QuerySelect From Strm

• Had better return Strm (It does)– Default [] window for Strm

– Default Istream for result


Simple Join QuerySimple Join QuerySelect From Strm, Rel Where Strm.A = Rel.B

• Default [] window on Strm, but often want Now window for stream-relation joins

Select Istream(O.orderID, A.City)From Orders O, AddressRel AWhere O.custID = A.custID


Simple Join QuerySimple Join QuerySelect From Strm, Rel Where Strm.A = Rel.B

• Default [] window on Strm, but often want Now window for stream-relation joins

Select Istream(O.orderID, A.City)From Orders O [[NowNow]], AddressRel AWhere O.custID = A.custID

• We decided against a separate default


Equivalences and TransformationsEquivalences and Transformations

• All relational equivalences apply to all relational constructs directly– Queries are highly relational

• Many low-level transformations in implementation

• Two new language-based transformations, more to come– Window reductionWindow reduction

– Filter-window commutativityFilter-window commutativity


Window ReductionWindow Reduction

Select Istream(L) From S [] Where C

is equivalent to

Select Rstream(L) from S [Now] Where C

• Question for audience – time to wake up!time to wake up!Why Rstream and not Istream in second query?

• Answer: Consider stream <5>, <5>, <5>, <5>, …


Window Reduction (cont’d)Window Reduction (cont’d)

Select Istream(L) From S [] Where C

is equivalent to

Select Rstream(L) from S [Now] Where C

• First query form is very common due to defaults

• In a naïve implementation second form is much more efficient


Some Subtleties (or Dirty Laundry):Some Subtleties (or Dirty Laundry):IstreamIstream, , RstreamRstream, and Windows, and Windows

• Example: Emit 5-second moving average on every timestep

Select Istream(Avg(A)) From S [Range 5 seconds]

Only emits a result when average changes

• To emit a result on every timestep

Select RstreamRstream(Avg(A)) From S [Range 5 seconds]

• To emit a result every second

Select Rstream(Avg(A)) From S [Range 5 seconds Slide 1 secondSlide 1 second]


More onMore on Istream Istream, , RstreamRstream, and Windows, and Windows

• Example: windowed join, emit new result tuples whenever there’s a new input tuple

Select RstreamRstream() From S1 [w1], S2 [w2]

• Istream may not emit results in presence of duplicates

• Rstream emits entire join result at each timestep or slide interval

• Correct solution requires union of multiple joins, somewhat messy


Some Windowing Rules of ThumbSome Windowing Rules of Thumb• Dstream used infrequently but is needed

• Istream and Dstream can be expressed using Rstream and other constructs — but don’t try it at home

• Istream always intuitive in presence of keys

• Rstream sensible with Now or Slide windows

• Row-based windows nondeterministic in presence of duplicate timestamps

• No need for partitioned time-based windows

(end of long aside)(end of long aside)


Filter-Window CommutativityFilter-Window Commutativity

• Another question for the audienceAnother question for the audience

When is

Select L From S [window] Where C

equivalent to

Select L From (Select L From S Where C) [window]

• Is this transformation always advantageous?


Constraint-Based TransformationsConstraint-Based Transformations• Recall first example query (simplified)

Select Sum(O.cost)From Orders O, Fulfillments F [Range 1 Day]Where O.orderID = F.orderID

• If orders always fulfilled within one week

Select Sum(O.cost)From Orders O [Range 8 Days][Range 8 Days], Fulfillments F [Range 1 Day]Where O.orderID = F.orderID

• Useful constraints: keyskeys, stream referential stream referential integrityintegrity, clusteringclustering, orderingordering


Evaluating a Query LanguageEvaluating a Query Language

• Is it expressive enough?

• Is it user-friendly?

• Can it be implemented?

• Can it be implemented efficiently?


Evaluating a Query LanguageEvaluating a Query Language

• Is it expressive enough? YesYes, see our Stream Query RepositoryStream Query Repository

Online auctions, network traffic management, habitat monitoring, military logistics, immersive environments, road traffic monitoring

• Is it user-friendly? To be determinedTo be determined

• Can it be implemented? Yes, see our demoYes, see our demo

• Can it be implemented efficiently? We think so, with lots of fun researchWe think so, with lots of fun research


Stream System Benchmark: “Linear Road”Stream System Benchmark: “Linear Road”

100 segments of 1 mile each

Main Input Stream: Car Locations (CarLocStr)

Linear City10 Expressways

Reports every 30 seconds

Reports every 30 seconds

car_id speed exp_way lane x_pos

1000 55 5 3 (Right) 12762

1035 30 1 0 (Ramp) 4539

… … … … …


Linear Road BenchmarkLinear Road Benchmark

• Primarily developed by Aurora group– With some help from us

• Suite of continuous queries based on real traffic management proposals. Example CQs:– Stream car segments based on x-positions (easy)(easy)

– Identify probable accidents (medium)(medium)

– Compute toll whenever car enters segment (hard)(hard)

• Metric:Metric: Scale to as many expressways as possible without falling behind


Easy ExampleEasy ExampleMonitor speed and segments of cars 1-100 Select car_id, speed, x_pos/5280 as segment From CarLocStr Where car_id >= 1 and car_id <= 100


Hard ExampleHard Example Whenever a car enters a segment, issue it the

current toll for that segment


Hard Example in CQLHard Example in CQL

CarSegEntryStr: Select Istream() From CurCarSeg

CurCarSeg: Select car_id, x_pos/5280 as seg, Location(expr_way, dir, x_pos/5280) as loc From CarLocStr [Partition By car_id Rows 1]

Select Rstream(E.car_id, E.seg, T.toll)From CarSegEntryStr [NOW] as E, SegToll as TWhere E.loc = T.loc


Hard Example (cont’d)Hard Example (cont’d)SegToll: Select S.loc, BaseToll (V.volume – 150)2

From SegAvgSpeed as S, SegVolume as V Where S.loc = V.loc and S.avg_speed < 40.0

SegVolume: Select loc, Count() as volume From CurCarSeg Group By loc

SegAvgSpeed: Select loc, Avg(speed) as avg_speed From CarLocStr [Range 5 minutes] Group By location(expr_way, dir, x_pos/5280) as loc


STREAM Query InterfaceSTREAM Query Interface


STREAM Query InterfaceSTREAM Query Interface


““Introspection” QueriesIntrospection” Queries• Real-time monitoring of query execution and

general system properties, for example:– Tuple-flow rate in a queue– Selectivity of a join– Ordering of a stream– Overall memory usage– Etc.

• Monitoring performed using regular CQL queries over extensible system-generated property streamproperty stream


Query Plan MonitoringQuery Plan Monitoring


For more information:– Talk to me this afternoon

– Visit our web site: http://www-db.stanford.edu/streamhttp://www-db.stanford.edu/stream

The Stanford STREAM Team: Arvind Arasu, Brian Babcock, Shivnath Babu, John Cieslewicz, Mayur Datar, Keith Ito, Rajeev Motwani, Itaru Nishizawa, Utkarsh Srivastava, Dilys Thomas

Documents

CQL: A Language for Continuous Queries over Streams and Relations Jennifer Widom Stanford University Joint work with Arvind Arasu & Shivnath Babu stanfordstreamdatamanager